RESERVOIR CHARACTERIZATION, AVO ANALYSIS AND QUANTITATIVE INTERPRETATION OF LOWER GORU, MIANO- KADANWARI FIELD IN MIDDLE INDUS BASIN, PAKISTAN

Abstract

Kadanwari Gas Field (KGF) is one of the prominent producing fields of Kadanwari-Miano block in Middle Indus Basin (MIB) or Central Indus Basin (CIB) and it was developed in 1989 by LASMO (London and Scottish Marine Oil) with gas reserves from Kadanwari-01 well at E-Sand of Lower Goru Formation (LGF). The research work aimed to assess the variability of reservoir regarding porosity, volumetrics, and saturations for the optimized productions in the producing field. The challenge arises in the characterization of gas prone facies from the rest of wet sands and shales due to heterogeinity in the reservoir and varible thickness. The distinction of producing facies is appraised by an integrated approach including seismic attributes, rock phyiscs modeling, AVO analysis, inverted elastic properties and petro-elastic relationships for volumetric estimations (water saturation, and porosity) through probabilistic neural networking (PNN). The data provided for this study comprised of 3D volume of post and pre-stack seismic along four wells i.e., Kadanwari-01, 03, 10 and 11 inside the volume. Pre-stack seismic data was in form of angle stacks (near, mid and far angle stacks) and all wells have the full suite of wireline logs, essential for reservoir evaluation, with exception of shear sonic (DTS) log, which was present only in Kadanwari-03 well. Seismic attribute technique illuminated the lithological content, structural discontinuity, geometrical feature, and hydrocarbon indication at reservoir portion in seismic section. The stratigraphic slices extracted at reservoir level indicated fractured sands with high instantaneous amplitude and low instantaneous frequency whereas fault discontinuity, necessary for hydrocarbon migration, was highlighted through coherence and semblance attributes. The high response of sweetness attribute pointed out the hydrocarbon presence and various attributes blended simultaneously through Red-Green-Blue-Alpha (RGBA) revealed the channelized sands. The hydrocarbon-bearing zone was differentiated within the E-Sand reservoir at well location through petrophysical analysis with 40% volume of shale (Vsh), 16% effective porosity (PHIE), and 30% water saturation (Sw), in producing well Kadanwari-01. In the study area, the heterogeneity is not apprehended through measured elastic logs, i.e., P-wave (Vp), S-wave (Vs), density, and their combinations, and identified litho-facies, such as shale, wet sand, and gas sands, DRSML QAU xxi could not be discriminated appropriately. The rock physics modelling rectified the measured log curves and remove the effects of borehole conditions, especially on density (RHOB) and modelled Vs (DTS) logs and produced optimized elastic attributes such as P-impedance (P-imp), S impedance (S-imp), and VpVs ratio. The process employed petrophysical properties and reservoir in-situ parameters as prerequisites to establish the petro-elastic models (PEM’s). The modelled log curves were consistent, concerning trends and normal ranges, with measured log curves i.e., P wave (Vp), S-wave (Vs), and density (RHOB) logs. These modelled elastic logs were used to established Petro-elastic relation using cross-plot between petrophysical logs i.e., PHIE, Vsh, and Sw and elastic attributes i.e., P-imp, S-imp, VpVs ratio and RHOB that discriminated the gas sand facies with low P-imp (8,000-9,000 m/s*g/cc) and S-imp (6,000-8,500 m/s*g/cc), reliable VpVs ratio around 1.55 and good PHIE (16%) with low Sw (30%). This petro-elastic relation was expanded over entire area using model based inversion (MBI). The stratigraphic slice extracted from acoustic impedance volume at reservoir i.e., E-Sand level provided the low P-imp in equivalence to the one established in petro-elastic relation at well location. The artificial intelligence (AI) based technique i.e., probabilistic neural network (PNN) was applied that dispersed the petrophysical properties over entire area and resolved the heterogeneity at reservoir level. The stratigraphic slices extracted at reservoir, differentiated the sand bodies from shale with higher response of PHIE against low response of Sw, their correlation with low acoustic impedance indicated the plausible location for gas sands. The pre-stack inversion provided additional information about lithology and fluid properties of the reservoir. The Seismic inversion extracted information is highly depended on quality of angle gathers, therefore poor-quality data with low signal to noise ratio provides compromising results. The provided angle stacks in the research were of low quality, however, the simultaneous pre stack seismic inversion (SPSI) was implemented to get an understanding of elastic properties at reservoir E-Sand level. The stratigraphic slices of inverted elastic properties extracted at E-Sand delineated low P-imp and S-imp with VpVs ratio and RHOB indicating potential zone. Lame’ Parameters i.e., Lambda-rho and Mu-rho were obtained through the inverted P-imp and S-imp established the Lambda-Mu-Rho (LMR) relation explained by Goodway (1997) for plausible location of gas sands. DRSML QAU xxii In provided wells, the DTS curve was accessible only in Kadanwari-03 well, initially it was calculated through empirical relation and then rock physics modelling was applied. Although the rock physics modelling provided optimized results when employed in the seismic inversion processes, especially post-stack, but for deteriorate seismic quality such as partial angle stacks, the best suited method of machine learning (ML) was adopted. The ML technique has been preferred in many fields including geosciences, for being robust, require less input data and high accuracy, to deal with data problems. The most used algorithms of supervised machine learning (SML) i.e., Random Forest regression (RFR), Decision tree regression (DTR) and Support vector regression (SVR) was applied in this research to get the maximum accuracy in desired data i.e., predicting missing logs and rectifying the effected log curves. RFR algorithm was proved to be more accurate with higher values of determination of correlation (R2 ) and low values of mean absolute percentage error (MAPE). The best algorithm i.e., RFR was further used for detailed reservoir characterization by generating the elastic attributes such as P-imp, S-imp, Lambda-rho (λρ), Mu-rho (μρ), as well as petrophysical attributes such as PHIE and clay volumetric (Vcl). The resultant attributes helped to establish a petro-elastic relationship delineated at the reservoir level. Possible gas zones were determined within high PHIE and low values of other attributes like Vcl, P-imp, and S-imp. The potential bodies in the form of channelized sands were also validated by low λρ cross ponding to higher μρ